The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill implements model and pipeline persistence using serialization formats that can be leveraged for arbitrary code execution if loading untrusted files. Evidence includes the use of pickle.load() in references/feature-engineering.md, joblib.load() in references/pipeline-orchestration.md, and torch.load() in references/training-pipelines.md.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits a vulnerability surface for indirect prompt injection due to the ingestion of external data files. Ingestion points: Data is loaded via pd.read_csv and pd.read_parquet in SKILL.md and several reference files. Boundary markers: No specific delimiters or isolation instructions are provided to separate untrusted data from the agent's logic. Capability inventory: The agent has capabilities for filesystem writes and network access to logging and storage services. Sanitization: The skill mitigates these risks by providing templates for data validation and schema checking using great_expectations.

ml-pipeline